Hey HN! We are Jonathan & Guy, and we are happy to share a project we’ve been working on. ChartDB is a tool to help developers and data analysts quickly visualize database schemas by generating ER diagrams with just one query.
A unique feature of our product is AI-Powered export for easy migration.
You can give it a try at https://chartdb.io and find the source code on GitHub.
Next steps ---> More AI.
We’d love feedback :) |
I dislike the multiple references to 'Magic' on the site, but I realise that's probably a subjective thing. Maybe I'm fatigued by the AI-in-everything trend.
I don't use all the DBMSes you support, but I'm not sure having a single query to run is really much of a selling point. For SQL Server, I'd rather execute a stored procedure with permission checks, and progress feedback, than a big chunk o' SQL. (Again, that may just be a me thing.) If it's an efficiency thing for very large DBs then I think you should emphasise that, and also detail how much faster it is to gather all the info in one fell swoop than if you split up the queries.
The Examples link is currently broken on your site, btw. That sounds like a really useful way that people can evaluate your application without having to run a query on their own DB, which they may be reticent to do for testing out a new app.
Two benefits coming to mind are that AI is a good buzzword, and "shove it into ChatGPT, that'll do" takes less effort than building a proper transpiler.
Unlike AI's non-deterministic techniques for example LLM approaches to NLP, CUE utilizes a well established alternative NLP deterministic approach namely Feature Structure [2], [3].
[1] Cue – A language for defining, generating, and validating data:
https://news.ycombinator.com/item?id=20847943
[2] Feature structure:
https://en.m.wikipedia.org/wiki/Feature_structure
[3] The Logic of CUE:
The chatbots are the Zero to One for programming.
So listen I agree with you. I’m not going to use this. But hundreds of thousands of smart people with time, like these guys, can now author stuff that will get better over time faster than you personally will solve any one particular problem for anyone. So something something, log in your own eye before the thorn in the other.
This would make Peter himself cringe let alone the rests of us.
That said, if you really believe this, I've got an AI product to sell you. It's the Zero to One for anything you want. It's called HypeLLM.
The OG of DB table designers is the SQL Server database diagrams feature, and before that MS Access, and there is still room for improvement in this tool. Obviously saving changes back to your db (and migration scripts) is the most important feature, git tracking of schema changes would be good.
> visualize and design your DB with a single query
> Instantly visualize your database schema with a single "Smart Query."
The tool seems really useful and I will give it a try!
Just curious about the emphasis on the single query. What's the value of this? If it makes it much faster than similar tools it's probably worth making it explicit. I don't use them often enough to know if speed is a big issue (if this is the reason).
Nice work anyway!
Marketing?
And they need a single json to create the diagram from.
I doubt the "single query" makes it immensely faster, but reducing query round-trips is usually time-saving. Might be noticable, depending on what you're comparing it to.
I just asked chatGPT for this, but human perspective would be great :)
Looks like not all databases (mssql in this case) like the generated result.
Not sure what kills it, but without "columns" in "full_json_result" it runs through.
can you elaborate on how the ai part works? im a bit confused how that fits in because there are many SQL diagram tools without AI as well
So you identified all the issues in your code generation, decided instead of fixing them to use ChatGPT instead, and trust that ChatGPT did it correctly?
Are you saving much effort? Are you providing an accurate, bug free experience for all the SQL databases you claim to support?
Congrats on the launch and happy building!
Disclaimer: I'm one of the cofounders of WhoDB.
This is true. In theory. In practice, the real struggle is to find the gem in an ocean of garbage.
Something that cannot be found simply doesn't exists for any practical terms.
Nature has a way to triage its sheer numbers of genetic permutations : wastness of the world and mostly infinite time.
No AI magic can know the foreign keys from the schema alone and even with data the results would be uncertain.
Column names could sound like foreign keys on pure chance and the data type doesn't show which other table it's referring to.
I’m 10 years in and have only ever seen it once